E-commerce · Guide 8 min read

AI chatbot for an e-shop: orders, products, support and complaints

In an e-shop a large share of messages repeat, and a fair number arrive in the evening or at the weekend. That is exactly where a chatbot makes sense — and also where the most can go wrong.

AI chatbot for an e-shop: orders, products, support and complaints
Author: Maksym Kamenchuk, Nexflow — Banská Bystrica Published Updated

A chatbot in an e-shop is most useful for three kinds of question: product availability and specifications, order and shipment status, and the terms for returns or complaints. All three repeat, and the answer can be worked out from data you already have.

What I do not recommend leaving to it: binding promises about delivery dates, individual discounts, expert advice on expensive goods, and the complaint decision itself. There the bot should prepare the material and hand it to a person.

From what volume it pays off: roughly from 30–50 repeated messages a week. Below that, answering personally is both cheaper and better.

What a chatbot in an e-shop handles well

Availability and specifications

Connected to the catalogue — the bot answers from real data, not from a description that has since changed.

Order status

The most frequent question in any shop. Once the order number is verified, the bot can answer immediately.

Delivery and payment

Options, prices, timescales and terms — typical content that is on the website but nobody looks for.

Returns and complaints

It explains the procedure and collects the material. A person decides on the complaint itself.

Help with choosing

With clear parameters (size, compatibility, power) it narrows the choice to a few items.

Capturing enquiries out of hours

When it cannot answer, it asks for a contact and hands over a complete record.

What not to leave to a chatbot in an e-shop

Binding delivery dates. “You'll have it tomorrow” is a promise the bot has no way of guaranteeing and that turns into a complaint. The bot should talk about usual timescales, not promise a specific day.

Individual discounts and prices. If the bot agrees a price, you will have to honour it. Leave pricing to a person or to fixed rules.

The complaint decision. Collecting material yes, deciding no. For legal reasons among others.

Expert advice on expensive goods. On a technically demanding purchase the customer values a person, and a bot can cause a bad purchase and a subsequent return.

Hiding that it is a bot. If a customer feels deceived, that is worse than not having the bot at all. Saying plainly “I'm an automated assistant, I can connect you to a person any time” works better.

Connecting to the shop: what matters here

A bot with no connection to data can only answer what is in its material. A bot with a connection can answer how things stand right now — and that is a fundamental difference in usefulness.

A security point that is often overlooked: a chatbot connected to orders is a new route to customers' personal data. Identity verification is not a detail — it is a condition of deployment.

When it pays off and when it is not worth it

SituationRecommendation
Many repeated questions, many outside working hoursa chatbot makes sense — the greatest benefit
A broad catalogue with clear specificationsa chatbot connected to the catalogue
Few messages, mostly during working hoursa chatbot will not pay back, answer personally
Expensive, technically demanding goodsprobably not — the risk of a bad purchase outweighs the saving
A messy catalogue, out-of-date stocktidy the data first, chatbot second
Nobody to take an escalationa chatbot only postpones the problem

A concrete payback calculation on your own numbers is in AI chatbot for business: what it costs and when it pays off.

When a chatbot pays offWhen a chatbot pays offSituationRecommendationMany repeated questions, many outside working hoursA chatbot makes senseBroad catalogue with clear specificationsChatbot connected to the catalogueFew messages, mostly during working hoursAnswer personallyExpensive, technically demanding goodsProbably notMessy catalogue, out-of-date stockTidy the data firstNobody to take an escalationA chatbot only postpones the problem
Diagram. Illustration: situations and recommendations from the article.

What to try before deploying a chatbot

Some questions arise because the information is missing from the website or cannot be found. These steps are cheaper and sometimes remove most of the reasons for a chatbot:

  1. Look at the last 50 messages and sort them by topic. It usually turns out three questions make up the majority.
  2. Add those three answers somewhere visible — on the product page, in the basket, in the confirmation email.
  3. Automate order status updates. An automatic email or message on a status change removes the single most frequent question entirely — and it is cheaper than a chatbot.
  4. Only then consider a chatbot for what is left.

This recommendation costs me work and I give it anyway, because a chatbot deployed against a problem a better email would solve is an unnecessary expense. Automated order status updates are part of e-commerce automation.

What to try before a chatbotWhat to try before a chatbot1 · Sort the messages• Last 50 messages• By topic2 · Add the answers• Product page• Basket• Confirmation email3 · Automate order status• Email or message on status change4 · Only then a chatbot• For what is left
Diagram. Process illustration: steps before deploying a chatbot.

Frequently asked questions

What can an AI chatbot do in an e-shop?

It handles three things best: answering about product availability and specifications, giving order and shipment status, and explaining the terms for delivery, payment, returns and complaints. Connected to the catalogue, it answers from real data rather than from a description that has since changed.

Can a chatbot access customers' orders?

Yes, but only with identity verification — for example a combination of order number and email, or a link from the confirmation email. Access should be read-only; a change to an order should be initiated by a person.

Does it work with WooCommerce?

Yes. WooCommerce has a REST API, so connecting to products, stock and orders is straightforward. Other platforms work too if they have an API or a regular export — I verify that first, before quoting.

From how many messages does a chatbot in an e-shop pay off?

Roughly from 30–50 repeated messages a week. What decides is the number of repeated questions, not the total message count or the size of the shop. How many of them arrive outside working hours also carries weight — those would otherwise wait until morning.

Can a chatbot promise a delivery date?

It should not. A binding promise of a specific day is something the bot cannot guarantee, and it ends in a complaint. A properly configured bot talks about usual timescales and hands the conversation to a person on a specific question.

Should the customer know they are talking to a bot?

Yes. Hiding it is a short-term gain and a long-term loss of trust. Saying openly that this is an automated assistant which can connect them to a person at any time works better in practice.

What should we try before a chatbot?

Go through the last 50 messages and sort them by topic. If three questions make up the majority, put the answers directly on the product page or in the confirmation email, and set up automated order status updates. It is cheaper and removes some of the reasons for a chatbot entirely.

Related services

ServiceE-commerce automation ServiceAI chatbot for business

Related articles

PricingAI chatbot for business: what it costs and when it pays off DecidingAI agent vs chatbot vs automation

Related work

Client projectE-shop product automation (WooCommerce)

Send me the last 20 messages from customers

No personal data needed, the questions alone are enough. From them I can tell you whether a chatbot pays off, what it should handle — and what you would be better off solving differently and more cheaply.

Get a free consultation →